Today’s episode features an interview between Matt Trifiro and Victor Bahl, Technical Fellow and Director of Mobility & Networking at Microsoft Research, and one of the fathers of Edge Computing. Victor's seminal 2009 paper titled “The Case for VM-based Cloudlets in Mobile Computing” spawned the vision for what we now call Edge. In this interview, Victor shares the story behind the genesis of that paper, and his experience of the evolution of edge in the years since. The Edge Computing World Conference is October 12-15th, 2020. Fully virtual. Register at www.edgecomputingworld.com and use the promo code OVERTHEEDGE for 30% off the Edge Executive Conference
Today’s episode features an interview between Matt Trifiro and Victor Bahl, Technical Fellow and Director of Mobility & Networking at Microsoft Research, and one of the fathers of Edge Computing.
Across a 23-year career at Microsoft Research, Victor has helped shape Microsoft’s long-term strategy through research, industry partnerships, and associated policy engagement with governments and research institutions around the world. His seminal 2009 paper titled “The Case for VM-based Cloudlets in Mobile Computing” spawned the vision for what we now call Edge.
In this interview, Victor shares the story behind the genesis of that paper, and his experience of the evolution of edge in the years since.
Key Quotes
“You can start to see a convergence of sorts between the communication and computing industry, and edge is the catalyst for that…Some sort of convergence is going to happen and it's going to be good because we are using each other's strengths and building on top of it rather than spending time doing the same thing.”
"The question for me as a researcher and a person who tries to think many years ahead was “what happens after cloud?”
“If you think about what the cloud is, the cloud effectively sells you compute and storage…and [we thought] wouldn't it be cool if the cloud not only sold storage, it not only sold compute, but it also sold latency...wouldn't it be cool to have an edge where the latency was small.”
“The internet works on a protocol called BGP, and that protocol has not been optimized for latency sensitivity. It has been optimized for cost sensitivity. So as the packets are routed, they are not routed with latency in mind. That’s why you end up seeing all these problems and jitter, which is a killer for many things.”
“If you actually create software, you can think of a network operating system. And then you can think of an API that you put on top of that. And then on top of that, you've got all these applications and then developers can party. So you're looking at a brand new ecosystem that is now seeded because of edge computing.”
“The way I like to think about it is “Wow, We're getting there! We've gotten there!” Because now when I talk to people, it’s no longer about trying to convince anybody of the need for edge. It seems like people get it…I no longer have to make the case for it. The thing that most people are looking at now is how to monetize the edge–how to actually light these things up.”
“I think that you can start thinking about edge as part of cloud even now. Because Microsoft is invested, we are building it out. We are building out the infrastructure. We are going to make it available. It's going to be there.”
Sponsors
Over the Edge is brought to you by the generous sponsorship of Catchpoint, NetFoundry, Ori Industries, Packet, Seagate, Vapor IO, and Zenlayer.
The featured sponsor of this episode of Over the Edge is Seagate Technology. Seagate’s new CORTX Intelligent Object Storage Software is 100% open source. It enables efficient capture and consolidation of massive, unstructured data sets for the lowest cost per petabyte. Learn more and join the community at seagate.com
Links
CLICK HERE to Register for the Edge Computing World Conference, October 12-15th, 2020. Fully virtual. Use the promo code OVERTHEEDGE for 30% off the Edge Executive Conference
[00:00:00] Matt: Hi, this is Matt Trifiro, CMO of edge infrastructure company, vapor IO, and the co chair of the Linux foundation's state of the edge project.
[00:00:07] Today, I'm here with Victor Bahl, technical fellow at Microsoft. We're going to talk about victor's background and career in technology his work at Microsoft research and his role as a pioneer in edge computing. Hey, Victor, how you doing
[00:00:20] Victor: today? Very good. Thank you very much.
[00:00:23] Matt: Yeah. So I'd love to start out by just asking you about, I mean, how you even got into technology.
[00:00:32] Victor: How did I get into technology? Well, like, millions of other,
[00:00:36]immigrants.
[00:00:38] I actually came to the United States from, India
[00:00:41] in when I was 18 years old.
[00:00:43] And, I,
[00:00:44] I started studying to be
[00:00:46] brutally honest about this, please.
[00:00:49] I was, I didn't even think about it. It just happened. So I
[00:00:53] Matt: checked the wrong box on your freshmen courses and liked
[00:00:56] Victor: it. I could make a nice story. I was like [00:01:00] twinkling with this and that, but the reality
[00:01:02] is I came to school,
[00:01:03] sounded like a good thing. Courses were good. I was good at some of these courses in physics and math.
[00:01:09] And so I had to say, Hey,
[00:01:11] this looks fun. So it just
[00:01:12] went on and on and on and put it. And so, yeah, I got into, I got another time when computers are not as prevalent, you know, and, It was, they were just about to start up. And
[00:01:23] so it was, there was a lot of excitement
[00:01:25] in the air. There was
[00:01:27] this new
[00:01:28] Victor: thing that we had, you know, which was very, so the, your imagination could run wild and think about, Oh my God, what are these things capable off?
[00:01:36] And what can we do with it? And so just one thing led to another. And so I just kept doing that, doing that, That's how it all began for me.
[00:01:44] Matt: Yeah. And you you've spent most of your career at Microsoft, is that
[00:01:48] right?
[00:01:49] Victor: Actually, I started my career, after I got my masters in, digital equipment corporation.
[00:01:55] So I don't know if you remember digital.
[00:01:58] It
[00:01:59] Matt: was,
[00:02:00] [00:02:00] Victor: yeah, it was a classy, classy company. Ken Olsen was the founder it was in,
[00:02:05] Massachusetts
[00:02:06] Victor: right next to MIT. and so when I got my first job there, and then at that time I had a job from Microsoft as well. Microsoft, at that time, it was a really small company, less than like 2000 employees.
[00:02:19] And, you know, they were looking at , which was, I don't even know if you remember that.
[00:02:23] Matt: Oh, I remember always too.
[00:02:24] Yeah.
[00:02:26] Matt: The cloud didn't even exist. The internet barely exists.
[00:02:30] Victor: Yeah, of course not internet wasn't a thing at the time. Not
[00:02:33] Matt: really, no. It was
[00:02:34] a research ARPANET,
[00:02:36] Victor: but deck a deck or digital deck, they had just brought in the queen
[00:02:41] Elizabeth to ship
[00:02:43] Victor: into the Boston
[00:02:44] Harbor.
[00:02:45] It was a. Big massive
[00:02:46] Victor: company. It had invented the workstations from, you know, going from
[00:02:50] super computers and stuff and it was on rock
[00:02:53] and roll. It was doing hardware software, the Unix operating
[00:02:57] system that was called Altrix.
[00:02:58] And I got a [00:03:00] job in a, in an advanced development group at the time.
[00:03:02] It was a, they were focused on image processing
[00:03:05] and video processing and multimedia the word, the multimedia didn't exist,
[00:03:10] Matt: but what rough year was this?
[00:03:13] Victor: Oh, so this was in 88. Okay. Yeah. So, so I decided,
[00:03:17] to take,
[00:03:18]offer at the time as opposed to Microsoft. And then I was there for about nine years
[00:03:24] before I joined Microsoft research.
[00:03:26] But,
[00:03:26] the
[00:03:26] interesting thing for my life was that, digital used to have this amazing program called Jeep.
[00:03:32]
[00:03:32] I think it was called a graduate education engineering
[00:03:35] program in which, if you applied and there was a big. Big process and, you know, lots of scrutiny and all that. But if you did, you could actually go and complete your PhD. and the company would take care of everything. All of the finances let's give you time off.
[00:03:52] It was
[00:03:52] great. I was the
[00:03:54] only one selected, and so that was
[00:03:57] quite a privilege. So I was there. So I got my
[00:04:00] [00:03:59] PhD at university of Massachusetts came back.
[00:04:02]then, in 1997, I think,
[00:04:04] I, I decided to leave that for various reasons. And then,
[00:04:08]Microsoft research was good for four years old at the time, five years old,
[00:04:12] four years, cause late 90 to start it.
[00:04:15] And, I started from there. And
[00:04:16] so I've been there since ever since. Yeah.
[00:04:18] Matt: Yeah. And you know, Microsoft research is, is actually pretty sizable. Microsoft spent a lot of money on, on research. but most people don't know about it, you know? And when I think of like deep corporate research programs, I think of like Xerox park, can you, for our listeners who may not be familiar with Microsoft research, can you just like describe what it is and how it relates to the core of Microsoft and, and all that.
[00:04:46] Victor: Sure. I would be glad to,
[00:04:48]it's for me to be in Microsoft research. So Microsoft research was set up by
[00:04:53] Nathan Myhrvold on behalf of
[00:04:56] bill Gates.
[00:04:57] So bill himself was
[00:04:59] a.
[00:04:59]you know, [00:05:00] bill is a
[00:05:00] fascinating human being and you've sort of seen no matter what people say about him, you
[00:05:05] know? but one of the interesting thing is that he, he wanted
[00:05:08] to have a research
[00:05:09] Victor: organization and he and Nathan and he, you know, got together.
[00:05:13] They invited Rick Rashid who was at that time at Carnegie Mellon university, a very well known
[00:05:18] professor. He had,
[00:05:20]Victor: build up the mock operating system,
[00:05:21] invited him
[00:05:22] Victor: to
[00:05:22] head up this organization
[00:05:24] Victor: that was around. 90 92 end of
[00:05:26] 1992.
[00:05:27]
[00:05:27] so Microsoft
[00:05:28] Victor: research was set up with three principles. One of them was to extend the state of art and the
[00:05:35] other one was to,
[00:05:36] Victor: keep, Microsoft viable in the future.
[00:05:39] But the number one thing was to extend the state of art.
[00:05:41] So
[00:05:42] Victor: we were built, the organization was built from ground up, not
[00:05:45] as a place to do research that just predominantly
[00:05:48] Victor: impacts the product,
[00:05:49] but research
[00:05:50] Victor: that. Predominantly impacts the entire community, the world. And so we were given the mandate to
[00:05:59] pursue [00:06:00] research
[00:06:00] in whatever we wanted to do, in the computer science field,
[00:06:03] math
[00:06:04] physics, all of that included and, feel free to
[00:06:07] publish these research results and papers
[00:06:10] are in the community.
[00:06:12] So we
[00:06:12] were set up like an academic department
[00:06:16]in, industrial research,
[00:06:18] community. The advantage for that was that we had a ready,
[00:06:22] set of problems to
[00:06:23] work with, that the world needed. And, there was no pressure
[00:06:27] put on us as researchers to impact
[00:06:30] the product right away.
[00:06:31] But if good things happen.
[00:06:32] Then great things
[00:06:33] would happen with the product.
[00:06:35] We would
[00:06:35] figure out a way to do that.
[00:06:36] So
[00:06:37] now Mike, somebody search now hosts about a thousand
[00:06:40] plus
[00:06:40] researchers all over the world as many, many labs from Bangalore to Beijing, to Montreal, to new England. And to Cambridge, UK Redmond and yeah, with,
[00:06:53] PhDs that are
[00:06:54] for people who have had PhDs and gotten from some of the top issues of the world,
[00:06:58] Stanford,
[00:06:59] MIT, [00:07:00] CMU, Berkeley, you name them.
[00:07:01] We get them with the personalities of the kind, which are very self driven. It, it is built.
[00:07:08]bell labs is a good, good example. Xerox park, you mentioned
[00:07:12] is a good example, right?
[00:07:13] I was a deck deck had a.
[00:07:15] SRC and CRL, which is Cambridge research lab and systems research lab centering. so, you know, we, we, we had people from there as well.
[00:07:24] Plus we had a new
[00:07:25] graduates plus we had some seasoned people
[00:07:28] and, the
[00:07:29] mix is beautiful because
[00:07:30] if you want to work in some space and you want to learn something about it, you probably have one of the top guys in the world working with you on
[00:07:37] Matt: that. That sounds like an amazing job.
[00:07:39] Victor: It's an amazing guy.
[00:07:40] It's a dream job. I have loved it. That is why I have not felt a need to move from
[00:07:44] there. I mean, I just
[00:07:45] invent, reinvent my job all the time without any shackles, but I do feel responsible to the company. So I do now, I do heavily motivated to use this trillion dollar trillion and a half dollar company to [00:08:00] get my ideas out in front of the world.
[00:08:02] And that's been a great.
[00:08:04] Matt: Yeah, that's, that's really neat. And, and, lots of things you've done in your career, you know, this is a show about edge computing, so I'm going to go right to that. I suspect that edge computing, was one of these reinventions, you know, you'll, you'll, you'll look at the, the, the literature and there's some, some early mentions of edge computing.
[00:08:22]you know, like, the, the AHCA, my guys, when they're at MIT, one of the first references of edge computing, but. In my mind, the, the, you know, the golden spike of edge computing was that 2008. I think I Tripoli research paper, which, which as I understand it, and doing research for, for this interview actually was something that was a derivative of the, this Epic meeting that you had in building 99.
[00:08:47] Can you tell me about that? Meaning and who was there and what was talked about and how this all came about?
[00:08:53] Victor: Sure. Sure. Thank
[00:08:55] you so much. It's very flattering that you've done. So I looked it up a little bit and you're you're right.
[00:09:00] [00:09:00] Tripoli
[00:09:01] Matt: paper personally was when the law that made the light bulb go off for me.
[00:09:06] Victor: Yeah. Yeah.
[00:09:06] So, yeah.
[00:09:07] Yeah. So thank you for that. The paper I believe was published in , computing
[00:09:13] magazine and it was in 2009.
[00:09:16] Yeah. The 2008, I think that you have is the time when we actually had that. Meeting. yeah. Let me tell you about that.
[00:09:23] So,
[00:09:23] you know, we didn't talk too much about my own research background, but one of the things that I have pushed quite a bit in my life has been, in the area of mobile computing and
[00:09:33] wireless communications and wireless networking and things.
[00:09:36] And so I, I built a part of my research areas by focusing
[00:09:39] on those, those elements or those parts of the field. So, I was
[00:09:46] around, and I helped, with the evolution or the creation of a cloud computing. In Microsoft. and our, the way I sort of did that
[00:09:54] was, we kind of understood
[00:09:55] that as the company moved from a shrink wrap software [00:10:00] company
[00:10:00] where you would just go to someplace and
[00:10:01] buy that CD and put it on your PC
[00:10:03] to the services
[00:10:05] Matt: or, or the 10, the 10 CDs,
[00:10:07] Victor: right.
[00:10:08] Matt: Just prior to the cloud, I think the last release of windows
[00:10:12] 10 CDs that's right,
[00:10:14] Victor: right. So we would need a.
[00:10:17] We would, everything has to, move into a massive
[00:10:20] infrastructure and really going to need a lot of networking for
[00:10:22] that. So
[00:10:23] we started working very diligently on,
[00:10:28] on some of that infrastructure piece,
[00:10:30] but then things were starting to go, you know, Microsoft is one of those companies that when.
[00:10:34] It sort of decides they want to go into direction.
[00:10:37] It has the power and the intellectual capital and people
[00:10:41] to do that.
[00:10:41] So then started moving really well.
[00:10:43] It was doing amazingly well.
[00:10:45] Now I sort of told you that I came from a world, which is wireless and mobile. And at that time, mobile. Oh, it was also taken off.
[00:10:51] If you remember, iPhone
[00:10:52] came out in 2007 and so it was really starting to take off.
[00:10:56] Matt: Yeah, I may ask you, you mentioned research, a lot of your research, was in [00:11:00] wireless. What, what, what sparks your imagination about wireless? What about mobile and wireless? What was it about about that? That excited.
[00:11:06] Yeah.
[00:11:07] Victor: So, So that is a, yeah, that's a
[00:11:09] interesting, so we can get back to that.
[00:11:10] I still got a minute,
[00:11:11] but, but that's fine.
[00:11:12] I think that's it. That's a sort of an
[00:11:14] interesting question. What sparked it?
[00:11:16] I think that,
[00:11:17] so
[00:11:18] as I was growing up, you know, I come from a family where my, mother, she's passed away,
[00:11:22] but she's a doctor.
[00:11:23] My father
[00:11:23] was in the government service, but they had a very service oriented mentality
[00:11:27] in terms of, you know, my daughter, my son, they all become doctors. So we have been brought up as well as have a very service
[00:11:33] oriented. So when I was in
[00:11:34] tech, One of the things I noticed was that the things we took for granted, like internet access, right.
[00:11:40] We just take it for granted, you and I are talking on this
[00:11:42] year
[00:11:42] of four or 5 billion people in the world who didn't have any of that. And they had no, no access.
[00:11:49] And it was just sad. It was just like, because
[00:11:52] we were creating this inflammation divide or digital divide, we were actually contributing to the, the, the [00:12:00] differential in economic, you know,
[00:12:02] Wellbeing of people and
[00:12:03] sort of, because if people don't have information, they cannot do, they cannot make good decisions and then they get left behind and
[00:12:11] more.
[00:12:11] So
[00:12:12] I observed that quite a bit and I. No, I wanted to
[00:12:15] do something for this
[00:12:17] and believe it or not, wireless seemed like the type of tech that could allow us to get this connectivity going at a price point where,
[00:12:28] Matt: well, you were very precious because that is exactly what happened. It turns out it's a lot cheaper to put a macro tower up than it is to dig up a bunch of streets or
[00:12:34] string
[00:12:36] wires between poles.
[00:12:37] Victor: And, and, and, you know, I I'm, I'm a realist. I understand like a
[00:12:41] business decisions are important. So
[00:12:43] people are not gonna like big, large companies are not going to pick up all these towers and things of that nature where there's no business. If people can't afford to pay it, they're not going to
[00:12:51] put this up.
[00:12:52] And so
[00:12:53] you had to build. tech
[00:12:55] in the unlicensed bands because, you know, if you've paid a
[00:12:58] lot of money for the frequency [00:13:00] license bands, you're going to want to recruit the cost back. But if it's unlicensed and you can build it in a way that the hardware is cheap, software is
[00:13:08] commodity or is open source or whatever,
[00:13:10] you could suddenly now have internet connectivity.
[00:13:14] So that's how I got started and why that's an
[00:13:15] Matt: amazing story. Yeah. I wouldn't have guessed that, but that's a great story.
[00:13:19] Victor: Yeah. So I wanted to do that a lot and I pursued that for a long time. before this edge happened, I, Craig Monday who was a
[00:13:25] CTO of
[00:13:26] Microsoft at the time, actually, talk to me about the duopoly, which was the cable and the telco duopoly at the time.
[00:13:33]And, you know, how we must, try to make sure that, the coverage was ubiquitous. Connectivity was ubiquitous. And so we took a two pronged attack to that. One was to hire people in Washington, DC, to do the lobbying with the government
[00:13:47] Congress
[00:13:47] to keep the prices down and keep it. And then the other one was, Hey,
[00:13:51] if you guys,
[00:13:52] don't become ubiquitous, we are going to become,
[00:13:54] so I took on the tech part of it, which is we sorta
[00:13:56] like build out a technology that, that could.
[00:13:59] William [00:14:00] Cortes threatening. The general
[00:14:02] telco is to say that, Hey, if you don't
[00:14:04] Matt: yeah. Time to accelerate. Yeah.
[00:14:07] Victor: And make it available everywhere. We are going to just go in and
[00:14:09] do it ourselves.
[00:14:10] Matt: So waiting for one of the cloud providers to buy one of the wireless companies, I won't ask you to signal what Microsoft's plans are, but yeah.
[00:14:19] It's certainly, it's certainly interesting. It's certainly interesting. And maybe just, maybe all you have to do is, is threatened, right? Maybe that may be the right answer.
[00:14:27] I actually, actually there is a
[00:14:29] Victor: actually it's not even about the threat or
[00:14:30] anything. It's, it's
[00:14:31] logical, right? If you think logically you
[00:14:35] have these three CS.
[00:14:36] Where do you make money? You have
[00:14:38] a computing industry at smaller fonts like Microsoft and even Amazon and
[00:14:44] Google and all these people. And then you have the content industry. Sure. Right. Which has Netflix and
[00:14:49] HBO. And
[00:14:50] Matt: would you play apps and in there,
[00:14:52] Victor: and you have communication apps? I would put in the computing industry.
[00:14:55] Okay.
[00:14:56] Okay. So, and then content is little content. I mean, just like Hollywood movies,
[00:14:59] you [00:15:00] know,
[00:15:00] Matt: TV, entertainment,
[00:15:02] Victor: consumption. That's right. And then you have the communication
[00:15:04] industry. So you
[00:15:05] have these three big, massive industry. I mean
[00:15:07] the community. Right. And so as you, I mean,
[00:15:10] I don't have to see anything if you just follow it along logically,
[00:15:15] you can start to see
[00:15:17] a convergence of Sox
[00:15:19] between the communication and,
[00:15:20] and computing industry.
[00:15:22] I mean edge, you know, we, we are going to talk about edge more, but edge is the catalyst for that. I mean, you and I are right now talking to each other. We are not using the telco infrastructure at all, but yet we are doing the things that have classically been in the
[00:15:36] operators and the carriers space. So.
[00:15:39] So, and we are investing a lot of money and they're investing a lot of money. There's a lot of duplication of effort. So right there, I mean, you know, like logically it makes sense, you know, few years from now,
[00:15:49] it's going to, some sort of convergence is going to happen
[00:15:51] and it's going to be for the good, it's just going to, it's going to be good progress
[00:15:55] because we are using each other's strengths.
[00:15:58] And then building on top of it rather than [00:16:00] spending time doing the same thing.
[00:16:02] Matt: Well, and you're already seeing some of the convergence, you know, you, you look at the mobile edge computing, and it's it's rebrand is multi-access edge computing because it turns out that most of the companies with wire lines have.
[00:16:14] Of a wireless wireless strategy, even if it's an envy and Oh, that's so yeah. And so they're going to use the 5g core to do the user provisioning and things like that. So that's yeah. Def deli faster. Okay. So let's get back to edge. I I'm glad we took that, that little side path. Cause that's really fascinating.
[00:16:30]so I believe the question that I, that I asked you was, you know, what was the, what was the, like, what, what. Made you think, or are we talking about
[00:16:38] the,
[00:16:40] you know, fix this? Or are we talking about the meeting? We were talking about the meeting.
[00:16:45] Victor: So, the way that I was going with that, when I was saying that, you know, I was working on cloud and
[00:16:49] infrastructure
[00:16:49] where I was going with that was that
[00:16:51] when things started to stabilize and
[00:16:53] I was saying, Microsoft is big.
[00:16:55] And you know,
[00:16:55] Microsoft is sort of,
[00:16:56]can get things done when,
[00:16:58] when they put their mind to it.
[00:16:59] So. [00:17:00] All that was going great. So, you know, from a research perspective,
[00:17:03] all that was going great.
[00:17:04] So now the question for me as a researcher and
[00:17:06] a person who sort of thinks
[00:17:08] ahead, you know, many, many years or tries to think ahead,
[00:17:10] was that what happens?
[00:17:12] What's after cloud, what is, what is cloud? And so.
[00:17:15] The observation. So what I did is at that time is, on that particular day, I actually remember the day it's October
[00:17:21] 29th. I remember it
[00:17:23] vividly. I invited someone, my colleagues, in particular invited, He just goes by there
[00:17:32] from Carnegie Mellon. Yeah.
[00:17:35] Oh yeah. Okay, great. And then I invited him Nigel Davis and writer,
[00:17:39] who is in the
[00:17:40] university of Lancaster. I invited Ramon. Sarah's not a hundred percent sure which company he was in at that time. And then, Roy, Juan, who is now in Intel, I had invited. And so what we did and the reason I know these, these guys.
[00:17:52] So if you think about like these, they were from universities and academia, right. And
[00:17:56] sorry, academia as well as industry. And
[00:17:58] the reason I invited them
[00:17:59] is [00:18:00] because,
[00:18:00] First of all, they were my cohorts. They, I mean, I knew them well, they are friends as well,
[00:18:03] but,
[00:18:04]
[00:18:04] we were, we used to think about these sort of things.
[00:18:06] Like, you know, what
[00:18:07] would we do next? So we invited them to Microsoft research and building nine cleared the entire day for me. And then, picked a conference room set there, got lunch, deliver breakfast to be delivered. And then we sat there and we stayed at the whiteboard and said,
[00:18:20] okay, What is, what is the next thing?
[00:18:24] And so, so it was a great day, of course. And, the thing that we really zeroed in on very quickly. And that came from because of her background was that if you well, the cloud and what really cloud is, the cloud
[00:18:37] sells you compute and storage. And of course it tells you all the services on top of all that stuff, but effectively surgery compute and storage
[00:18:45] and a diamond.
[00:18:47] Again, we realized a few things. Once we, because we were from the mobile world, we said, we knew that phones
[00:18:53] are resource limited.
[00:18:54] Right. They don't at that time, the phones didn't have, I mean, even now,
[00:18:58] Matt: even that little battery is the [00:19:00] big limiter now
[00:19:01] Victor: battery and, even GPU, if you want to do rendering
[00:19:04] of,
[00:19:05]they didn't have a tremendous,
[00:19:07]
[00:19:07] sort of resource like that, disconnections used to happen, right.
[00:19:11] Network is, was not like always
[00:19:12] connected.
[00:19:13] So what do you do with that place? And then the other thing was latency, right? Like there were all these applications of research being
[00:19:20] an MSR. I was
[00:19:22] privy to all these amazing researchers. They are working on some fabulous,
[00:19:26] like holographic stuff, right? Like you could actually, in fact that researchers are doing
[00:19:31] hello.
[00:19:31] Hello teleportation.
[00:19:33] So you can actually see somebody sitting next to you, you know? And so anyway, so, how is this
[00:19:38] ever going to become real. Right because this acquired,
[00:19:41]latencies, which were
[00:19:42] really low.
[00:19:44] And so,
[00:19:45]I,
[00:19:45]as we
[00:19:46] discussed more and more, it became clear that wouldn't it be cool if the cloud not only sold,
[00:19:51]storage, it not only sold compute, but it also stored latency.
[00:19:56] But
[00:19:56] the problem was that
[00:19:58] internet latencies are [00:20:00] uncontrollable.
[00:20:01] Matt: Right. Right. When you're, depending on a bunch of other, other people's best effort networks, it's hard to deliver an SLA.
[00:20:07] Victor: That's right. And so again, think about it logically
[00:20:11] and then it sort of made sense. Yeah.
[00:20:12] Okay. The computer camp, if you have the data can come to us, we're going to go to the data.
[00:20:17] So. Well, you got to create these
[00:20:20] subsistence to be like,
[00:20:21] people had been talking about things of this nature about ubiquitous computing world,
[00:20:26] about embedded systems
[00:20:27] or embedded things in their environment and things of that nature. So we thought wouldn't it be, wouldn't be good to have edge where
[00:20:34] the latency was small.
[00:20:35] So if it was a wifi connection and a, for some other connection, which is less than.
[00:20:39] One millisecond latency
[00:20:40] from the play, from the way the data was originated to where it has to be processed, then you would have it. So
[00:20:46] that's the big idea that can, and then of course the, when we, when we hit that idea, then we started with saying, okay, well, this is great for disconnected operations, too.
[00:20:55] Because if, if the cell phone thing
[00:20:57] goes away, you can still operate. So you have disconnected off on you. [00:21:00] So
[00:21:00] you know that if you just look at the IOT world, now, one of the
[00:21:02] things that sells most is in places where they don't have great
[00:21:06] connectivity.
[00:21:07] And so they have edge computing to actually keep the processing and things, going to the connectivity, come back in fact,
[00:21:12] yeah.
[00:21:13] Matt: Form of edge computing, different than on premises computing.
[00:21:18] Victor: So, on-premise while it's not that different, I mean, on-prem is, not that different, although I would say that,
[00:21:25] edge computing people,
[00:21:26]you know, this, very well, so, Different people think of edge computing differently.
[00:21:30] So, you know, there's, the whole problem was about the definition of what that even means. So I actually used to, not object, but get irritated by that
[00:21:39] sort of way. Cause I
[00:21:40] had a certain definition in my head, but then, as being a researcher again, I started to realize that it
[00:21:45] was fine.
[00:21:46] Like let it led to tremendous amount of exploration on prem computing was less about edge per se, but
[00:21:54] just about the fact that you didn't want the data to move out
[00:21:56] of where you were and you wanted to compute, she'll be closer. You want it to feel [00:22:00] secure about that
[00:22:01] edge is a lot more about
[00:22:03] a person useless computer offloads, for example,
[00:22:05] you know, you, you mentioned , you know, buying it and did a lot of work with CDN, but.
[00:22:12] They weren't not at CDN was just a caching thing.
[00:22:15] Matt: Data offload.
[00:22:16] Victor: Yeah. But there wasn't,
[00:22:18] there was no processing per se, because
[00:22:20] once you decide on processing, you go down, that's a different path altogether.
[00:22:24] Yeah.
[00:22:25] Matt: I added some processing later with the sort of workers concept, but it was really about doing some last minute, you know, rendering type stuff.
[00:22:32] Victor: Yeah, yeah, yeah. So, so yeah.
[00:22:35]computing became started to become a thing. I have a story for you
[00:22:39] about how it hit,
[00:22:40] if you're interested in.
[00:22:42] Matt: Absolutely. I'm interested. The history of it's fascinating.
[00:22:44] Victor: So I actually, got in front of the SLT, Microsoft SLT.
[00:22:48] Steve bomber was the CEO at that time in
[00:22:50] 2010.
[00:22:52] And I,
[00:22:52] presented these ideas that,
[00:22:55] you know, this is the next big
[00:22:56] thing I said,
[00:22:57] you know, we've got to, whoever gets the [00:23:00] latency, war is going to win
[00:23:01] because you're going to just enable all these applications that didn't exist in the past.
[00:23:06] So Steve, at that time asked me a question that I wasn't prepared for.
[00:23:10] He
[00:23:10] said, what would you change today?
[00:23:12] Okay.
[00:23:13] So I, you know, I always was thinking of the future,
[00:23:16] Matt: right?
[00:23:17] Victor: You asked me how my
[00:23:18] future, I would tell you,
[00:23:19] Matt: and this is 2009.
[00:23:21] Victor: 2010, 2010. And I'm in front of the SLP, which means all the, you know, at that time, I don't recall EVP, but basically the top press
[00:23:29] Microsoft, including the CEO
[00:23:31]you know, 15 of them or 20 of them.
[00:23:33] And I'm
[00:23:33] standing in front of him and telling him what didn't change today. So
[00:23:35] now I'm like, okay. So I just mumbled something like,
[00:23:40]speech, because
[00:23:40] speech Martin, my head speech models were very big at that time. And, you know, so I thought, eh, instead of going to the cloud, you would bring it to the edge.
[00:23:47] And so things would go right. And there was an EVP who, who managed all the speech thing and.
[00:23:53] He
[00:23:53] took offense to that or what I
[00:23:55] worry, you know, he's sorta like,
[00:23:57] no. And so, anyway, there was more, [00:24:00] more fun in that meeting. I wouldn't go to that because
[00:24:03] that's in itself.
[00:24:04] But however, I kicked myself or not having a
[00:24:08] tremendously good answers, clear answers at that time.
[00:24:11] So what happens is,
[00:24:12] so as usual, great
[00:24:13] research, interesting ideas, but you know, asking for two months, like, you know,
[00:24:17] cause they would have to spend a
[00:24:18] bunch of money.
[00:24:20] So anyway, I go back then in 2014, I get another chance.
[00:24:25] So that's four years
[00:24:26] later, and this time I nail it, like completely nail it,
[00:24:30] like
[00:24:30] in the sense that it was a 30 minute talk, went on for one hour and 30 minutes, they everybody's engaged.
[00:24:36] The CEO is
[00:24:36] engaged. He hasn't he's engaged asked
[00:24:38] me questions because this
[00:24:39] time I was systematically walked through, not just the idea, but all these different things that I had, bear it out and showed them how this would be
[00:24:46] so much
[00:24:47] better. And it sort of opened their eyes
[00:24:48] like. The kinds of things I showed at that time.
[00:24:50] So I came back. So he, he asked me Steve, at that time, how it's going to cost.
[00:24:54] And I
[00:24:54] was like, you know, again, I wasn't prepared. I don't know what you mean.
[00:24:57] But Satya who
[00:24:58] was sitting in that meeting, he was [00:25:00] not an EVP at that time. He was not a CEO at that
[00:25:01] time. He actually said something like $380 billion or something.
[00:25:05] He said, And Sue said done,
[00:25:07] and I'm like, what,
[00:25:09] what is this? Like,
[00:25:11] I
[00:25:11] don't know. I didn't go in there with like, anyway, but things don't work this way. Like, it's not like, you know, in an engineering companies of
[00:25:20] the Microsoft caliber, it works different
[00:25:22] next. Okay. So the CEO is all bought in. Others are bought in.
[00:25:25] So now you have to work with the absolute technical top.
[00:25:28] Breasts, right.
[00:25:29] I wasn't a technical fellow at that time. So they are distinguished engineers and technical for these are the top people in the technical area in
[00:25:36] Microsoft. And so
[00:25:39] they started saying, I'm an, I met with them afterwards because now they had to actually get right.
[00:25:43] And there were people who were not naysayers, but they said, Hey, every bird, the cloud is going to be within 30, the milliseconds of everybody on the planet. If that is the case, then why do we need edge? Right.
[00:25:55] That's kind of what
[00:25:56] I said,
[00:25:57] obviously said the usual
[00:25:58] things that you hear now quite [00:26:00] a bit, but Hey, you know, words are cheap in Microsoft.
[00:26:05] Corn is King, you kind of barely got to show up. So at that point, I sort of realized that. Man, no matter how much I say this is our word battle, because I'm asking for too much in, from a perspective of how much build out daily going to need. And, you know, there
[00:26:18] are all these questions to which you may come to, which is a management.
[00:26:21] And how do you manage edges? You got a truck for a problem, et cetera.
[00:26:26] Anyway, long story short, I went back determined training. You need to show them where the money is
[00:26:31] in some sense. And so,
[00:26:33] as I was thinking about it, I was in a sabbatical that led me to,
[00:26:37] the live video analytics things. Cause I saw cameras everywhere
[00:26:40] and I thought, and I thought to myself,
[00:26:42] what the hell who's looking at these cameras?
[00:26:43] Right.
[00:26:44] You know, like
[00:26:44] who's looking at the live feeds
[00:26:46] and there isn't was that speak to, and
[00:26:48] things were bad. And
[00:26:49] so it was always like after the
[00:26:50] fact,
[00:26:51] right. Something bad happens and then go back and look at the camera.
[00:26:53] And I thought.
[00:26:55] We have the cloud. We have computer vision, you know, we have every element.
[00:26:59] Now we [00:27:00] have the edge.
[00:27:01] That's it.
[00:27:02] I need to show that, you know, cameras are going to be everywhere. They're going to be part of our life
[00:27:07] anyway, for different security, purpose,
[00:27:09] congestion management, traffic,
[00:27:11] all kinds of good stuff.
[00:27:13] And I'm going to use it on the edge.
[00:27:15] Yeah. So I started pursuing that.
[00:27:17] Matt: Well, it turns out cameras are good for things you've been outside of the visible spectrum, which we don't tend to think about temperature reading with infrared and things like that, which might be really, yeah.
[00:27:27] Applicable given COVID right now.
[00:27:29] Victor: Yeah, that's right. That is exactly right. And, and, you know, people sometimes, but privacy, no, I actually thought about privacy
[00:27:37] from
[00:27:37] first principles and I, you know, I, I wanted to build everything that was, I would never compromise
[00:27:43] human privacy in that sense. And,
[00:27:44] Well, we can talk about that some of the time, but nevertheless,
[00:27:48] that was a
[00:27:49] first and foremost in my head when I was thinking about cameras, but I knew that
[00:27:52] cameras can be used for tremendous good too.
[00:27:55] So I started working on that, pursued that pretty hard heavily. And now, by
[00:27:58] the way, you can sort of see [00:28:00] that Microsoft just announced a
[00:28:01] live video analytics product that is out there. Other companies announcing it,
[00:28:06] we open source some of that stuff
[00:28:08] and it all works on the edge. But in the meantime, as we were doing this IOT hit.
[00:28:13] And that happened because it was a
[00:28:15] VC.
[00:28:16] I think his name was
[00:28:16] Anderson.
[00:28:17] He gave a talk and said the cloud is dead. It a very provocative, doc cloud. Is that, so everybody like Peter Levine, Peter Levine. That's right.
[00:28:25] Matt: Yeah. I know Peter really well. He funded my last company.
[00:28:27] Victor: Yeah. Okay, great. Wonderful. So it's at that attention.
[00:28:32] Yeah. I think our
[00:28:32] CEO and everybody else.
[00:28:33] And so.
[00:28:35] What does it mean? But what he was seeing was that in manufacturing, for example, industry 4.0 in other places, retail and other places that, you know, you needed to have some computing, everything couldn't be moved to cloud. And so we had also, acquired Lincoln, I mean, at that time.
[00:28:50] So Kevin Scott joined us as a
[00:28:53] CTO. He was very much on top of this stuff. He
[00:28:56] really understood it. He got
[00:28:58] to the, to the ear of [00:29:00] Saudia, explain to him why this is important. And then Satya Mader. And they made an
[00:29:04] announcement. If you're gonna spend
[00:29:06] $5 billion in IOT and edge. Well, that had a ripple effect on the entire company.
[00:29:11] And I died a nation because when a company like that is going to put $5 billion, that's a lot that just shows you there's a market,
[00:29:19] you know, that's a big market. And so now you started to see.
[00:29:23] A lot
[00:29:23] more things start to happen.
[00:29:24] So while I was on this, a journey on video analytics, I still believe that annex is a truly killer app, but IOT hit pretty big.
[00:29:31] And so everything. Yeah. So that's kind of how
[00:29:34] it works. I'll stop here cause I'll be in touch.
[00:29:35] Well, no, and I,
[00:29:36] Matt: and, and I've, I've been, you know, as a marketer, part of my struggle has been, how do we explain its computing to people that are coming at it fresh that don't understand it. and one of the things that I've started talking about, and we published it in the last day of the report is the three acts of the internet.
[00:29:53] You know, the first act was, it's just amazing that. You can sit in a browser and connect to any computer anywhere in the world that was [00:30:00] on the internet and request something. And the fact that it may have taken, you know, hundreds of milliseconds or even ones of seconds was fine because it was just, just pure magic that you could actually get it.
[00:30:10] And then we tried to start consuming, you know, high resolution video, and we got the buffer symbol and that's when the CDN really became, because we were mostly downward consumption. This was back when we had DSL and you know, it was. asymmetrical, you know, you got a lot more downward bandwidth and we wanted to consume these large images and these rich media websites and videos without buffering, and the CDN saw that, and that was sort of the second the internet.
[00:30:32]but now we're in the story, which I believe edge computing is the defining component. And the biggest difference to me. And I'm interested if, if you also see this is that we're moving from a world of primarily humans talking to machines, or maybe humans talking to other humans. Like we are now, which again, we tend to operate in one sub-seconds right.
[00:30:52] I mean sure. You know, frame rates and stuff, fractions of a second, but in general, our conscious. Our conscious minds, operate in ones of [00:31:00] seconds to, to a machine that's glacial. And we have billions of machines all generating yeah. You know, terabytes of data because they can, what, what are you you gonna do with all this?
[00:31:13] And, it seems pretty clear that there's a lot of really actionable information that has a half-life of. Tiny amounts of time. And, and so, so I've, I've really seen it moving into this world. Like what's going to create this like demand. And I think IOT really, it's another way of looking at IOT. It's like, we're moving to a world where you can have all these machines talking to other machines, putting out data that is going to be, w lost most of its youth, by the time a human ever has a chance to see it.
[00:31:41] And you've got to have this like low latency. So let's talk about latency a little bit. So you mentioned like 30 milliseconds help, help our listeners understand like what, what that, where that 30 milliseconds come from. So what that means, and then. how it's achieved today.
[00:31:56] Yeah, yeah. Right. So I, I,
[00:31:58] Victor: Julie, [00:32:00] by the way, I want to comment too, on the way you've described it.
[00:32:03] I think you're right on, that, we are going to need computing elements close that, and then the operation speeds is way beyond, humans, comprehension
[00:32:11] and human speeds. And so.
[00:32:13]and we can't take time to move that data to the cloud, but given the internet as exists. So then having an edge computing device, which can actually, build this machine to machine
[00:32:22] communication and machine to machine processing
[00:32:24] is going to be very, very
[00:32:26] important as we move forward.
[00:32:28] So,
[00:32:28] are you right on it? the 30 minutes, second thing number that I gave you sort of thrown at me as, like how. much time, you know, as we continue to build out more and more cloud data centers around the world, from, different people than the amount of time it takes for packets to go leave, leave your device and
[00:32:47] get to the,
[00:32:48] Matt: so to put that in kind of like terms that people understand all over the world, there are these Azure zones.
[00:32:54] Right. These, these, these essentially data centers that either you've built our own, or maybe you're leasing space from [00:33:00] somebody else's facility, but it's a it's someplace where Azure has compute. and there are so many of them, if I'm understanding correctly that for most people with a. Wired connection because as you know, wireless is a we're in the process of fixing wireless.
[00:33:15]well the wired connection is, is 30 milliseconds or less from one of those data centers. Is that correct?
[00:33:21] Victor: But, I, I want, yes, but it's going to be even less than 30 milliseconds. So our aspirations is our, you know, that was a while ago that the conversation
[00:33:29] happened as have actually,
[00:33:31]Become even more.
[00:33:33] And we, as we are building out a and we build a lot more and more you're correct. And everything, which is that the wireless wired
[00:33:40] connection to
[00:33:41] getting to the first point of
[00:33:42] entry to as, or will be less than 30 milliseconds, for example, things like that.
[00:33:48] Matt: and that counts the time over the last mile network, which is typically owned by somebody else.
[00:33:53] Victor: The last, well, it depends. So if it's a wireless, then it doesn't count that first cause of wireless. I'm an LT. The rule
[00:34:00] [00:33:59] of thumb, I use 70 minutes
[00:34:02] anyway. So, but yeah, for the point to generate a battery in a wide network and you send it
[00:34:07] to us, we are pretty close to you that way.
[00:34:09] Right. But here's, here's the deal.
[00:34:11] I mean, so, you know, you, you're trying to sort of, Get it to the point where users will understand what does that mean? Really?
[00:34:16] That's what I was going
[00:34:17] Matt: for.
[00:34:17] Victor: We know about seconds. And then 30 minutes
[00:34:19] seconds. I mean, that doesn't seem to
[00:34:20] like long. And why is that a product? Isn't that enough? The thing is that the way a network protocols are designed, they are very chatty.
[00:34:29] They have a lot of overhead, you know, and, and for good reasons, it's not like, you know,
[00:34:33] the design.
[00:34:34] Not smartly. They were designed for particular
[00:34:36] reasons. I mean,
[00:34:37] Matt: Eric correction that can arrive out of order, all kinds of resilience in it. Yeah,
[00:34:41] Victor: that's correct. Yeah, it works. And then congestion happens, lots of packets in the air.
[00:34:45] You need to back off, you got to like figure it out and you need to play nice.
[00:34:49] So anyway, so,
[00:34:50] so that 30 milliseconds, which seemed like a real small number,
[00:34:54] actually.
[00:34:55] Starts to compound. When you think about the time it takes right. Packer to [00:35:00] deliver. So it will be great if like a packet starts to deliver and then boom, boom, boom.
[00:35:03] Everything goes, but it doesn't happen that way. Right.
[00:35:05] It's
[00:35:06] just a, it's just a number. And then if I could show you. We'll make it visual. It will be much more easier. But if you think about gaming as an example, and you sort of think about how much the blank yeah. How much time it takes for their package to get there.
[00:35:19] It has to render, it gets to decompress the packet, render rate processes, coming back, and then you add, keep adding those milliseconds. And on top of that, you add 30 milliseconds on the add all the roundtrip latency,
[00:35:30] TCP or
[00:35:31] UDP or whatever, a packet. Oh, the protocol you're falling. That number then starts up.
[00:35:37] Looking very big and gets to the point when the gaming is not great. Cool. like when you,
[00:35:42] you're doing first person shooting game or some other,
[00:35:45] you know, live game, you cannot tolerate those sort of
[00:35:48] agencies at all.
[00:35:49] So 30, 30 minutes seconds see a big number. And then going back to what you were suggesting in a machine to machine communication, it's an eternity.
[00:35:58] I mean, that is just
[00:35:59] not kind of [00:36:00] work.
[00:36:00] So really we need to come down to one or two milliseconds and a, and there has been a lot of work. I mean, you hear about five G a
[00:36:07] lot, you know,
[00:36:08] sort of. But in five G for example, it's been in progress for a long time and people I've done the analysis of how much latency,
[00:36:16] different applications can, tolerate.
[00:36:18] So if you
[00:36:18] think about AR VR, you know, how much can it tolerate if you think about platooning where we could, we could falling.
[00:36:23] So you have congestion on the roads I can follow, and it turns out.
[00:36:28] But usually
[00:36:29] it is less than 10 milliseconds.
[00:36:31] So if, if we can give them a network or if they B, which means if we can give them a way to
[00:36:36] get their packets from the time of generation, to the first computing device
[00:36:40] within 10 milliseconds and magic can happen.
[00:36:43] And so that's kind of how you do all the calculations. That's why, you know, networks are built now that have been built that way to provide you with this new
[00:36:50] class of applications.
[00:36:51] So these milliseconds matter a lot.
[00:36:54] Matt: Yeah. And one of the things that, that, I didn't realize until I studied this field for a while, while, [00:37:00] you know, we talk about latency and we talk about roundtrip latency, but a lot.
[00:37:04] A jitter is actually in many cases more important because when you're trying to control something, you know, a robotic drone or robotic arm, something that, that you need the discrete predictable latencies below a certain threshold. And it's one thing to say, 10 milliseconds on average, and you might have a tail, you can see that's 90 milliseconds, you know, that happens once a day, which.
[00:37:25] Might be fine if, if it causes a hiccup and fortnight, but it might be a disaster if it's, you know, providing decision support to a vehicle or something.
[00:37:33] Victor: That's correct. Yeah. Predictable networking. Yeah. It's very
[00:37:37] predictable.
[00:37:38] Matt: And, and you know, the two things that, that cause latency and jitter are. Is the well, the, the, the distance right.
[00:37:45] Speed of light becomes important at longer distances. And then the number of network hops. And it's pretty intuitive that the farther out in the network, you get that the more of those she reduced. And as you, as you mentioned, you know, you've got, you've got the entire internet between you [00:38:00] and the traditional cloud.
[00:38:01]
[00:38:01] so if you look at that, let that, let let's say,
[00:38:04] Victor: and then we haven't even talked about wireless. Like if you have wireless to the mix, Then it's a different ball game together, right? It is. Yeah. And internet is not like the internet works on a protocol called BGP or border cable protocol. And that protocol has not been optimized for
[00:38:19] So it has been optimized for cost
[00:38:21] sensitivity. So
[00:38:22] as the packets are outed, they are not routed with the latency in mind. And so that's why you end up seeing all these problems and gender, as you mentioned, which is killer for many things.
[00:38:32] Yeah.
[00:38:33] Matt: So, so if you, if you look at it at this general function of edge, and let's say it's, it's sub 30 milliseconds, but you're right, there's a sort of inflection point at these, you know, 10 milliseconds.
[00:38:42] There was actually an interesting one at sub one millisecond. You know, if you look at the, the, the virtualization of the five G networks, you know, the, the, you know, the distance between the remote radio head and the, the, the BBU is. 75 microseconds. Right? So just, just below one millisecond. and that's about 15 [00:39:00] kilometers.
[00:39:00]so, so two questions. One is, how do you get compute? Farther out than it is today. Where, where does it go? And then who owns it and how do I consume it? I do I do I buy devices and put them on yeah, my premises or do I consume them from a cloud provider like Microsoft? And if I'm consuming it from a cloud provider of Microsoft, where are those machines, you know, are they in micro data centers, adjacent to the base band units and the cable head ends?
[00:39:25] Are they on premise with like an Azure stack? Is it a mixture of both? Like how do you see that sort of thing? Rolling out. Where, where do you see that? How do you see that happening?
[00:39:34] Victor: Yeah. Yeah. So there's actually a lot of information in the question that you've asked me. and so let me, let me talk. Is this the part of it?
[00:39:41] I've got the
[00:39:42] Matt: expert on the line. I'm going to
[00:39:43] try it out as much as like, yeah. Oh, this
[00:39:45] Victor: is, this is great. This is great. I love the conversation. so yeah, you, you put a lot of information in that question.
[00:39:50]So let's
[00:39:52] just back up and think about Fiji. I brought it up and you brought it up. you mentioned the 30 millisecond.
[00:39:57] I mentioned 30 minutes. Second, you picked up on that. [00:40:00] But again, I want to emphasize that that was many, many years ago. Now
[00:40:05] when aspirations is
[00:40:06] to be even less. So, you know, I don't want to give a number per se, but nevertheless it's lower than 30 milliseconds. but anyway, going back. So I think the way, 5g being and the way it was described, been belt, you understand there's a mobile core, there's a radio access network part of it.
[00:40:22] And there's obviously provided a network part of it. And all of that has
[00:40:26] to work and everything has to be pretty low latency.
[00:40:29] And so, so I think the trick is going to be so, 5g, for example, has a concept of network slicing. They have a, they have a
[00:40:38] term which is called. If I get,
[00:40:39] I'll try to reliable Linklater their
[00:40:41] communications,
[00:40:42] which is where they specify one to four millisecond on air latency, right?
[00:40:47] From. From mobile devices and package
[00:40:50] to the base station. Now
[00:40:51] Matt: it be sub 10. You've only got another six milliseconds to
[00:40:55] Victor: that's. Right. And it's not just six months. You'll have to process that.
[00:40:58] So
[00:41:00] [00:40:59] you'll get you get RF signals, you know, and then you've got to move them and you go back into, is them made into IP packets to all the mobile core.
[00:41:06] So mobility has been all that good stuff. So your time budgets are pretty low.
[00:41:10] So
[00:41:11] which is why edge is starting to. Take off in a huge way in the operators and telecommunication world as well.
[00:41:18] So
[00:41:19] the interesting thing to me is that, a lot of the functions that these, that the operators used to do are done in hardware, right.
[00:41:27] They were done in, very specialized pieces of
[00:41:29] Matt: hardware appliances.
[00:41:31] Victor: Yeah. But that was a problem too, because if you sort of think about how long it takes to, for us to go from one generation to another
[00:41:38] generation, it takes tens of years. Cause it costs billions of dollars. Yeah.
[00:41:42] The interesting thing is that I, that I had mentioned earlier was that now.
[00:41:47] Well, you believe you can do all this in software. I mean, there is,
[00:41:50] Matt: well, I mean, you open up an Erickson boxer, Nokia box, and it's an Intel PC.
[00:41:55] Victor: It is these in, for specialized hardware anymore. We can do all of it.
[00:41:59] Matt: Yeah. [00:42:00] cGAS and some that you can even deal on the, on the, the core Intel, the high end Intel chips.
[00:42:04] Yeah.
[00:42:05] Victor: Right. And then if you do it in software, where are those boxes sitting? That's
[00:42:10] Matt: the question. That's where I was getting at.
[00:42:12] Victor: Yeah. So then, we'll see how the world evolves. I mean, you know, all the operatives have, central offices so I think, we did some golf places in, and, I think I read in the spec too.
[00:42:21] You can go actually up to 40 kilometers away from the base stations. but 20 kilometers is for LTE that you have, that you can
[00:42:28] be away and still be able to manage
[00:42:30] Matt: from the, from the. Between the, the remote radio head and the base station from the base station to, yeah.
[00:42:35] Okay. Correct.
[00:42:36] Victor: And so, so then, the question is that,
[00:42:39] okay, so real real estate
[00:42:40] becomes interesting year now who owns the real estate.
[00:42:42] Right?
[00:42:43] And if the opera didn't have the real estate and you know,
[00:42:45] Microsoft has a real estate
[00:42:46] or some other company has stayed in there, we can put computing there. That's when a computer is going to be and that's
[00:42:52] okay. So as
[00:42:53] you look at the future and you sort of think about like
[00:42:56] your users are people, you
[00:42:58] know, who are thinking [00:43:00] about, okay, so I understand you guys figured out the low latency stuff, but how does it help
[00:43:03] us?
[00:43:04] So envisioned the falling world. Now, if everything in software can actually create
[00:43:09] abstractions, you can create an equal
[00:43:11] and a wind 32. So if I may give you an analogy of the PC, all right.
[00:43:16] So the PC,
[00:43:17]came about, there was hardware. Microsoft wrote the operating system. I'm an Apple leader too, but make sure the operating system.
[00:43:24] And then on top of that created with 32 API and then the world party,
[00:43:28] right. There was so many companies that got
[00:43:30] built fortunes were made
[00:43:32] PCs everywhere.
[00:43:34] Now the network. If you, if you were actually create software, I mean, if it's all about software, you can think of a network operating system. And then you can think of an API that you put on top of that.
[00:43:44] And then on top of that, you've got all these applications that developers can party. So you're looking at a brand new ecosystem that is now seated
[00:43:54] because of edge computing. So
[00:43:56] there is this classical notion of edge computing that we've [00:44:00] conceived a long time ago, which was about putting computing.
[00:44:02] Close to the, to the, where the data is generating. But the newer version of computing is where the edge computing is part of the infrastructure and it's actually abstracted out so that now you can use you're much more close to the Island or network in this
[00:44:17] particular gears and started. And so
[00:44:19] now if you go down the path of either IOT
[00:44:21] applications, machine to
[00:44:23] machine, or do AR VR or vehicles or
[00:44:26] whatever, you can start to say,
[00:44:28] well, I don't actually have to exist in the cloud.
[00:44:30] I can
[00:44:30] access on the edge. I can exist somewhere
[00:44:32] else. And so I think that's how the world will evolve and sort of. Job makes that available to people faster and makes it easier to them. I love Microsoft because Microsoft has done this,
[00:44:44] like this sort of stuff
[00:44:46] in the past quite a bit. And because we have, we love the developers.
[00:44:50] We build a lot of, software to make their
[00:44:52] lives easy. So
[00:44:54] at Microsoft has all the elements that. Yeah, I can make it
[00:44:57] exceedingly successful in this quest [00:45:00] to,
[00:45:00]as we embrace the new world and then
[00:45:02] just change the way people think about, where the computing sets and, you know, a
[00:45:06] way to write the
[00:45:06] applications and development and things of that nature.
[00:45:09] Matt: Yeah. And I, you know, and I, I, I come from the cloud world and so infrastructure is relatively new to me in this idea that like the cloud exists somewhere physical, and it has moving parts. It's got fans and things like that. And when you really get into this world, like you wonder why anybody. Other than a cloud provider would actually ever want to own a server.
[00:45:29]data centers are messy, places that are hard to run. it's hard to make them efficient, unless you've got some economies of scale. It's a really interesting world to be in very
[00:45:38] Victor: capital
[00:45:39] intensive business, not
[00:45:41] Matt: capital intensive, and it's
[00:45:42] Victor: not an
[00:45:42] easy business because
[00:45:44] you know, there is churn on
[00:45:45] these servers.
[00:45:46] We have to constantly buy new servers, constantly buy new hardware, constantly
[00:45:51] fix the software.
[00:45:52] And why would like, for example, if you,
[00:45:55] if you have a store, which,
[00:45:56] you know, you're selling ice cream or you're, you know, doing salad, but you still need some ID [00:46:00] or your dentist's office, you know, why would you have to, I want it all
[00:46:03] Matt: sounds dumb terminal,
[00:46:06] Victor: but then at the back end, you're right.
[00:46:08] It is not, it is not like. Easy or work. It is
[00:46:12] clearly work in some sense, it takes, it's
[00:46:15] very hard. And to run these things on scale, I mean, you need, you need actually pretty smart people who understand the science
[00:46:22] of it or can
[00:46:23] build it and keep it together and make it work all the time. So, yeah. Yeah.
[00:46:27] Matt: And you know, let's go back to one of the original comments you said where you said, you know, well, we, we sell, we sell storage, we sell compute.
[00:46:34] Why aren't we selling latency? And, Today I can provision an Azure instance in one of 60 or so, zones. Is that right? Something like that. and I can get presumably get storage in those same zones. I can imagine a future where I could request a low latency network slice in one of those sounds.
[00:46:52]so let's imagine that that world exists that, and let's imagine it's not 160 it's 2000. Right. Cause you're, you've got them in the central [00:47:00] offices and these micro data centers that are adjacent to the base fan units and all this stuff. where is the software for managing that? Right. You know, in the, in the old world where I could, I was mad.
[00:47:11] Yeah. Managing, you know, four data centers in one country. I could. A human could, could, could allocate those things. But when I'm worried about running a workload or having a piece of data where I need it, when I need it at a particular time, maybe at a particular cost, because ed resource constraint, they might be auctioned based like where is the, the.
[00:47:33] Where's the software in the abstraction layers to make that possible. Where, where are we on that?
[00:47:37] Victor: Yeah, that's a that's. That's great. You, I mean, yeah. How do you scale on this, this level? So,
[00:47:43]Matt: even right. A basic app that's, that's so highly distributed, but it's actually,
[00:47:49] Victor: so I think, I think, Right.
[00:47:50] There are two again, two questions here. So one is actually off as a management question, which is that. So all cloud companies, not just, Microsoft, [00:48:00] have figured out a way to, and have some
[00:48:02] very
[00:48:02] sophisticated management software, which. Insurance that everything, all their servers are running well, and they do predictive analytics to ensure if something is going bad, they can actually bring it down or bring something back up, back it up.
[00:48:15] And so that consumer and the user never sees any differences. So similarly, application writers, developers. you know, they, they get a VM and they run, they run their programs on it and then things go bad. We can actually move things around fast enough that they don't see the difference. So now okay. So that word, we have invested quite a bit
[00:48:33] on.
[00:48:33] And so
[00:48:34]all the difficult, I think we, we do pretty good on
[00:48:36] it. Now we understand the tech, we built it
[00:48:39] from ground up and
[00:48:40] we are constantly improving it.
[00:48:42] So in this new world, now you have
[00:48:43] Matt: Victor, you're your audio?
[00:48:45] , So we know
[00:48:45] Victor: pretty well how to manage a large infrastructure with millions of servers.
[00:48:51]and we manage it very well in this new world, that you and I just talked about where, we have actually already gone and we have edge zones and we have edge [00:49:00] devices and creates computing devices that are at the
[00:49:02] outskirts of
[00:49:03] the large clouds. But
[00:49:05] inside the now inside.
[00:49:07] They're inside our networks sometimes, sometimes, outside the network, we have been building management software that allows us to
[00:49:14] take into consideration
[00:49:16] the speed of the network, the mothership, and these edge devices.
[00:49:21] And then, and that is still able to recover if any failures happens and false happen. So we are going to get to the point where we will manage this. And then of course, we will figure out how to scale this. Well, The second question you asked was, from a developer's perspective, where does this all set in?
[00:49:38] Where are the applications? And they're all over the world and all over from thing. Can you do it? That is a more interesting problem to me. I mean, there is,
[00:49:46] there is,
[00:49:47]I think there is work to be done, quite a bit. I mean, so right now, if you think about it, we, in the cloud where we offer these
[00:49:55] notions of regions, right?
[00:49:56] So Microsoft, for example, has 54 regions. And so you can say [00:50:00] U S West.
[00:50:00] I want my applications to work on the
[00:50:02] data centers that exist in the U S West, which covers it.
[00:50:05] And so in some sense, it's like you prescribed a little bit and then inside a U S West, there are multiple data centers then, you know,
[00:50:11] you can, we can control
[00:50:13] where it goes and
[00:50:14] manage the latency
[00:50:15] in this world.
[00:50:17] You, in the edge world, you may have to be a little prescriptive, either prescriptive, or you will have a monitoring systems that can actually predict what latency you're getting from different
[00:50:27] teachers in different ads zones, based on that latency.
[00:50:30] Could move your,
[00:50:31] computes into those areas, right? So
[00:50:34] we've actually got in research, we've
[00:50:35] got some projects going on, which, which do that, which sort of says,
[00:50:39] if I have a packet coming, which ed should it hit.
[00:50:42] And from that edge,
[00:50:43] which data center it should hit.
[00:50:45] And, since there are many, many options or many edges that I could hit and it sort of depends on. what is the latency to the edge? What is the sort of the, how much the edge is being used? You know, free to now, which how do you want to go? Which data center you want to go.
[00:50:59] So it's a joint [00:51:00] optimization problem. That one. It has to come up with, so those are the kinds of technologies we are actually building, or we have in our building. And then from our developers, our goal would be that the developer had, doesn't have to think too much about this.
[00:51:13] Matt: Some declarative
[00:51:15] Victor: specification, right.
[00:51:17] They gave us that's right. And then we, from that, we can figure it out and we can build
[00:51:20] the underlying systems so that we
[00:51:21] can move things around seamlessly. And provide them try to do the best to give them the,
[00:51:27] the SLA that they have accounted for.
[00:51:29] So these are the amazing innovations that'll happen over the next several years and things would become,
[00:51:35] a much, much more simpler.
[00:51:36] I D thing is to keep it as simple as possible for
[00:51:40] anybody to be able to use this infrastructure piece. Yeah. And so, you know, the, the,
[00:51:45] Matt: the, the origin of, of all this edge can sorta be traced back, you know, to that, that meeting in 2008, it's
[00:51:53] Victor: 2020.
[00:51:54] Great. Yeah.
[00:51:55] Matt: That's 2020, like, right. That feels like an eternity.
[00:51:59]
[00:51:59] yes.
[00:51:59] W [00:52:00] where are we on the edge journey? And when, when are we going to reach critical mass? Like, how close are we.
[00:52:07] So
[00:52:11] Victor: one way to think about it is the way you put
[00:52:12] it,
[00:52:13] which is that okay. That was like, 1112 years ago, man. And like what the heck, you know, the other way to worry about it, which is the way I like to do think about it. Wow. We're getting there. We've gotten there because now when I talk to people, it's a slightly different problem that I see.
[00:52:30] So it's not no longer about trying to convince anybody the need for edge.
[00:52:33] It seems like.
[00:52:34] People get it.
[00:52:35] Matt: Well, that's happened in like the last two years.
[00:52:38] Victor: Yeah. Yeah. So they get it and that's okay. You know, rule of thumb that I've used in my world of research has
[00:52:43] been seven years
[00:52:45] for us. It takes seven years showing from the point where you conceive an idea in a lab.
[00:52:50] To the time that it might see a light lighter day or, you know, get out there and that's sort of average, right? It can go more. It can go last show when you get to like 10 years or eight, nine [00:53:00] years. It's not like so much far away from what I thought this would happen because, because even though like things seem to be moving very fast, the pace of innovation is
[00:53:08] very
[00:53:09] fast, but it takes time for big ideas
[00:53:11] to seep themselves in.
[00:53:13] They are they're disruptive. And the cloud was a big idea.
[00:53:16] Edge is a big idea. So
[00:53:19] anyway, so going back, when I talk, so, individuals everywhere, I, I no longer have to make the case for it.
[00:53:24] It's
[00:53:24] not even a thing anymore. No. The thing that people, most people are now looking at is how to monetize the edge.
[00:53:31] Right. How to actually a light up these things, but yeah, certain texts that we are waiting for, like the rollover of five G, but we are waiting for five mg and then that, as that comes actually becomes a lot more. Useful in the industry, in the industrial
[00:53:46] sector, manufacturing space
[00:53:48] industry, 4.0 is all about infusing it into a manufacturing process
[00:53:53] to improve efficiency.
[00:53:54] So that is happening too in their edge is sort of the deep pain that that is making it all [00:54:00] happen. So I believe you asked the question, when do you think it's cool? I think that we are, the momentum is increasing. I think that. Over the next several years, the pace of innovation, the pace of products. Like if you think about large companies and the number of products that are coming out, they're quite good and quite impressive.
[00:54:18] I mean, some of the products that you can now buy on the edge. And, so I, I would say, that in a
[00:54:24] few years you should see, you know,
[00:54:27] I only actually believe it's already there, but you should see a lot more pervasive.
[00:54:33]So when, thing that I kind of want to say, Okay. So, let me just turn to a
[00:54:40] five G for a second. Right?
[00:54:42] Matt: I wasn't asked about five G because we have wired connections that can deliver super low latencies. Why is an edge computing here with wired connections?
[00:54:52]Victor: because I think the
[00:54:54] most interesting things in this world
[00:54:56] that need computing resources are mobile and [00:55:00] sensors.
[00:55:00] Yeah.
[00:55:01] I mean, I, I literally feel that, I mean, even when you talked
[00:55:04] Matt: about the internet of mobile things and the internet of things that are hard to put wires to, cause there's so many of them.
[00:55:09] Victor: Yeah, and they're going to be so pervasive and they're going to be billions of CIS devices. I mean, you can't even imagine, like I've been actually, they have, colleagues in academia all over the place.
[00:55:18] I've been sort of working with them. one front of mine is in Princeton and he's talking about putting. These wires inside the homes like in the wall. So they actually act as antennas. And then, you know, they, they improve propagation. There. They are people talking about
[00:55:32] it. So sensors, these mobile sensors
[00:55:35] and these mobile device, these devices are going to be
[00:55:38] sober wasted.
[00:55:39] It's gonna light up the world in a
[00:55:41] really, really great way.
[00:55:42] And so for that, you need wireless. and that
[00:55:44] needs edge too,
[00:55:45] because they just, they need there. As you mentioned earlier on the show, you talked about battery and
[00:55:49] energy consideration
[00:55:50] and things of that nature. We need to keep all of that in mind.
[00:55:53] Matt: Well, in fi five G is, is, as you pointed out, it's, it's just building out the five G [00:56:00] infrastructure itself, especially with the. The virtualized network function is creating demand for compute at the edge to run the
[00:56:08] network.
[00:56:09] And so it's not that big of a leap to say, well, gosh, the marginal cost of adding a few racks of servers to run application workloads is small compared to the cost we had to do to just upgrade the network.
[00:56:21] So I totally see that. I see how that's driving a tipping point. That's that's a really interesting perspective.
[00:56:25] Victor: Yeah. I, I think of wireless as a computing problem. I don't think a wireless is a communication problem and I've worked in wireless all my life.
[00:56:32] Matt: Right. It's really interesting. Have you always thought of it as a competing from
[00:56:36] Victor: no, actually I've started thinking about it more recently about it as competing problem.
[00:56:40] And I'll tell you what I, what I mean by that. What I mean by that is that. Now, for example, you've heard of my memo. You've had a massive, my mode terms
[00:56:49] for the users.
[00:56:50] I mean, multiple and multiple add port multi-user memo is
[00:56:53] when you actually can form beams to watch every client and you can actually simultaneously send packets to them.
[00:56:57] And so you can
[00:56:58] pass
[00:56:59] basically
[00:57:00] [00:56:59] Matt: enables orders of magnitudes, more devices to simultaneously,
[00:57:03] Victor: right. And increases the capacity
[00:57:05] from a same spectrum that you
[00:57:06] have. Right. I think what Shannon limited then you think about it. Okay. Anyway. So the point is that. We can, we can now, because we do everything in software, we can increase the capacity of your
[00:57:17] spectrum
[00:57:19] Bible by the computer that we provide
[00:57:21] to it.
[00:57:22] So if you give me signal, I'll give you lots and lots of signals. I can actually disambiguate from that signal and get useful information out of that. And that's all computing problems. So by, by tying the communication with the computing, I am starting to now handle the more fundamental problem or spectrum unavailability or not having enough spectrum.
[00:57:43] I don't know if you remember, like several years ago, there was a lot of like, who would cry, you know, lots of, like, there was a presidential committee
[00:57:50] about spectrum being a scarce resource and we don't have enough and all that kind of stuff. Right.
[00:57:54] That's true. Fundamental. But. But nevertheless, have we done enough with it?
[00:57:58] We actually have some [00:58:00] amazing technology and information
[00:58:01] theory information systems,
[00:58:03] and we can now light all that up because now we are looking at that as a computing problem, and we already know how to deal with scaling up, compute how to put more commodity servers in there, how to put more containers in there.
[00:58:14] We know how to sort of paralyze stuff. And so while this is going to be like, wow, now we should not have a capacity problem. We should not have a disconnected problem. We should not have congestion.
[00:58:24] We should
[00:58:24] not have, you know, all the other problems that we have. And I was just going to be like a phenomenal thing that will make all of that happen.
[00:58:31] And Oh, by the way, you're totally right.
[00:58:34] Even the operating system on the
[00:58:35] edge will kind of have to be real time-ish, but then there will be enough CPU leftover, which the application
[00:58:41] writers developers can use
[00:58:43] when it's not doing
[00:58:44] that other stuff. Right. So now you can say, well, I could monetize that hardware computing, the
[00:58:48] general purpose computing that I've made available at the edge.
[00:58:51] I can offer this to them on a person by
[00:58:53] God. That's going to be
[00:58:54] awesome. I'm going to make a lot more money. And then developers are thinking, Oh wow, I have a new way because now I can not, I don't have to worry about [00:59:00] latency. I can put this. Gaming applications, this telemedicine stuff that I've been wanting to do this machine to machine thing I've been wanting to do.
[00:59:06] So
[00:59:06] all that gets hot and enabled.
[00:59:09] So yeah.
[00:59:10] So
[00:59:10] Matt: a win, win. When do you think that the edge is just going to be part of the cloud? We're not going to talk about any differently. It's just going to be, Oh yeah, of course you can provision a low latency workload in, you know, East Chicago. when do you see that point where it's just the way we build applications.
[00:59:28] Predicting future, what can get, you can cheat and say seven years. Yeah. Looking
[00:59:34] Victor: at the crystal ball. what can I tell ya? I think, I think we were, I hope we would not include Kobe when we were talking about this. I would be a lot more optimistic right now. It's still very optimistic. I think we are going to, we've got to come out of all this stuff that is happening to us right now and into, you know, waiting again and doing some great stuff.
[00:59:54]I th I think, well you could start thinking about edge as part of cloud even now. [01:00:00] Wow. Like, cause we are invested in Microsoft is invested, you know, we are building it out. We are building out the infrastructure. We are going to make it available. It's going to be, there, the operators, you know, are interested.
[01:00:11] I know they are very, very interested in, in, in. Working with us and enjoying because they benefit tremendously, you know, it's crest upside for them. It's a beautiful man.
[01:00:21] Matt: Oh yeah. New revenue streams. Well, yeah, it's someday. And I don't know, I don't know when this is going to happen. but it's conceivable that, that one of the telcos would say, well, why do I even want to own servers?
[01:00:33] Why don't I just run my core network functions on. Microsoft servers, for instance, you know, from that. But I could see that I could see, I could see a competitive, just a different capital structure for a telco.
[01:00:46] Victor: Yeah. Yeah. I mean, we did, this is public knowledge, so there's nothing major I'm telling, cutting about, but, you know, we acquired a firm networks.
[01:00:54] That's
[01:00:54] Matt: right. Yeah.
[01:00:55] Victor: Networks is the number one provider of mobile core
[01:01:00] [01:01:00] services. So yeah. So,
[01:01:03] yeah, you're not that far
[01:01:04] away from the thing that,
[01:01:05] you know,
[01:01:07] do you think this is, this is
[01:01:08] great for everybody. I really do. I think that we need to, build on each other's strengths and, you know, we are strong in certain areas and the, and the operators are strong in another area.
[01:01:19] And so, yeah, I think joining forces is the natural thing to
[01:01:22] do. It's
[01:01:23] really, really good to do so. So you're not too often. You're. Thinking about what is going to happen and what might happen in the
[01:01:30] future. Yeah. The Victor that is a
[01:01:32] Matt: great place to wrap. your enthusiasm is infectious. I really, really enjoyed talking to you.
[01:01:38] I've followed your work since I discovered it, which was probably three or four years ago when I got into edge computing and, it's helped me come to a new understanding. And it's, it's, it's fun to hear how your perspective has changed. And I hope some day to have you back on the show and we can see what's what's happened between now and then.
[01:01:55] Victor: Oh, it's my pleasure. Thank you so much for your enthusiasm as [01:02:00] well. It's really nice to see
[01:02:01] somebody who was
[01:02:01] as enthusiastic and knowledgeable as you are, and has been in the, in the industry and looking at it deeply. So thank you for the great questions. I, I had a good
[01:02:09] time too, and I hope some of your people who hear this, so we'll enjoy it as well.
[01:02:15] So thank you.
[01:02:16] Matt: How, how can people find you online?
[01:02:19] Victor: Oh, I'm very easy to find. I mean, any go to any of your search engine, just type my name, but.
[01:02:24] Matt: You mean the Bing search engine, right?
[01:02:26] Exactly. Of course.
[01:02:28] In the edge browser,
[01:02:29] Victor: the big search engine in the edge of, is there something, is there anything else?
[01:02:34] Of course, of course.
[01:02:35] Yeah. It was
[01:02:36] Matt: an absolute pleasure.